Shenzhen
Joined August 2023
家人们俺搓了一个行研研究,数据采集和质量检测的AI Agent! 可以收集市场上的产品数据,快速分析insight,检测信息是否正确/有无ai幻觉 自带金融终端接口,统计计算,电商数据,学术检索,高级图表功能 可用于金融分析,行业扫描,学术研究等需要统计+数据研究的领域! 完全免费家人们!
1
406
vibe coding出了一个数据分析的agent,可以画复杂的图表,找长尾数据,做统计计算和分析,还可以画麦肯锡级别的PPT。 可以用来预测市场,分析趋势,做投研,做策略模拟和回测等。用来预测世界杯哪个队伍能赢。
1
130
基模是deepseek v4 flash,其实做下来发现只要tool和agent框架足够强,最后呈现的能力并不好比顶尖模型差。 但确实在对“概念”的理解上会略微逊色,但实际上大部分任务并不需要如此哲学细致般的思考(opus4.8常丢上下文中英文不清也是这个原因),只要把tool和逻辑做好就能达到优秀的效果。
107
great step of crypto
Agents execute on Uniswap We've released seven new Skills giving structured access to core Uniswap protocol actions Your starting point for agentic workflows onchain
1
282
Ziggy Chen retweeted
《SAAS 创业八个月,我有点不信 Build in Public 了》 创业八个月了,我开始怀疑那些关于"Build in Public"的主流说法。 一开始我完全相信这套理论。开通了小红书、公众号、Twitter,甚至在社交媒体上记录整个创业过程。这些建议听起来很有道理:分享你的过程,早期获得反馈,围绕产品建立社区。 但现实是什么样的呢? 来的人根本不是客户——而是其他创业者、工具收集爱好者,还有创业圈的围观群众。每条更新都吸引同一批人:想要免费试用的程序员、寻求认同感的准创业者,甚至还有来做市场调研的竞争对手。 当你在公开场合构建功能时,给你反馈的不是那些会为产品付费的人。你得到的是"专业早期用户"的评论,他们的主要兴趣是免费获得新工具的使用权。关键是,你根本分不清他们的热情来自真实需求,还是仅仅因为能免费尝试新东西而兴奋。 在这些社区里的对话让人沮丧。大家想讨论宏大愿景和市场理论,而不是我的产品试图解决的具体痛点。他们会建议一些听起来很厉害但与付费客户实际需求完全不符的功能。这完全就是披着用户调研外衣的学术自慰。 我开始潜伏在其他创业社群里观察,发现不只是我遇到这种情况。到处都是同样的模式:活跃参与者主要是开发者、想创业的人,还有我称之为"专业测试员"的群体——这些人像收集宝可梦卡片一样收集新工具的试用权,但从不成为付费客户。 这个认知让我如醍醐灌顶:在公开创业空间最活跃的人,往往最不能代表你的真实市场。 如果你要服务的是主流用户——而不仅仅是技术圈的早期采用者——那么为Product Hunt发布和创业Twitter病毒传播而优化可能是适得其反的。你本质上是在用那些永远不会付费的人的反馈来训练产品开发流程。 拿我的AI应用举例。我想触达的是普通职场人士,而不是AI发烧友。但猜猜当我分享更新时谁在主导对话?是那些想讨论模型架构的AI极客,而不是只是希望内容创作流程能提升10%效率的忙碌营销经理。 残酷的真相是,Build in Public经常创造一个回音室。你最终会和其他也在尝试公开构建的创始人对话,形成这种奇怪的元社区,大家都在表演创业,而不是真正服务客户。 我学到的是:来自公开社区的早期反馈通常是噪音,而不是信号。真正的客户通常忙于自己的实际工作,没时间在创业社区里给创业者提供免费咨询。 愿意参与你的公开构建过程的人,往往落入几个可预见的类别:做市场调研的竞争对手、探索副业项目的开发者、创业观光客,以及可能有也可能没有真金白银的投资人。这些群体都没有你真正目标客户的紧迫性和约束条件。 我不是说Build in Public完全没用。对某些特定类型的产品——特别是开发者工具或专门针对创业社区的产品——它可能有效。但如果你要服务主流市场,你可能完全是在为错误的受众优化。 真正的问题不是Build in Public是否帮你找到用户——而是是否帮你找到正确的用户。经过八个月的尝试,我越来越确信,对大多数产品来说,答案是否定的。 也许是时候承认,最好的客户调研发生在无聊的地方:行业论坛、LinkedIn群组,以及与真实目标用户的咖啡馆对话——而不是在创业Twitter和创业者抖音的光鲜世界里。 有时候最有价值的反馈来自那些永远不会关注你公开内容的人,但如果你把他们的问题解决得足够好,他们会默默掏出信用卡。 这是我希望八个月前就有人告诉我的道理。
57
37
24
329
87,693
obviously Soham is a good guy, but this story has a little bit funny
There’s a lot being said about me right now, and most of you don’t know the full story. If there’s one thing to know about me, it’s that I love to build. That’s it. I’ve been isolated, written off, and shut out by nearly everyone I’ve known and every company I’ve worked at. But building is the only thing I’ve ever truly known, and it’s what I’ll keep doing. Earlier today, I signed an exclusive founding deal to be founding engineer at one company and one company only. They were the only ones willing to bet on me at this time. The team is cracked, they back misfits, and they’re building something absolutely insane in the video AI space. We’re launching at the end of this month. More details tomorrow in my TBPN interview. I’m pissed. And I’ve got something to prove. Soham
1
447
Ziggy Chen retweeted
my workflow now is short prompts with simple constraints i do not expect a perfect answer back every time. i cede control to the models to let it be more creative. i @ add context to give it clues. the back and forth human-ai dance feels the most fun to me
16
4
74
10,964
有意思,上下文工程比单纯的提示词工程有更多的“信息源”和“规则”支撑 底层模型越来越强,对于语义理解的泛化能力也飞速提高,提示词反而作用不大 但是一个可供后期维护的结构,真实可靠的数据等,对于让agent完成任务来说都是必要的 提示词是chat bot时期的东西,上下文工程像是一种任务导向的需求
在深度开发 LobeChat 的时候我就发现单纯的 Prompt Engineering 可能不再能很好地表征 llm 应用的实现。「 Context Engineering 」 这个说法,的确更好表达了当前的 AI 应用的本质
2
5
2,647
😋坎耶韦斯特特特特,你怎么回事了了了
This quoted post is unavailable.
186
Ziggy Chen retweeted
how it feels to write code without AI now
141
368
77
8,506
505,107
翻到两年前的帖子,还挺有意思。 现在大家说的创造一个结构化的,可以良好操控llm的prompt自然语言,通过语言学的概念和规则也可以做出来,可能还需要加些编程语言的规则,如套嵌,对象。 不过语言本身就是一种知识,现在可以用llm得到以前不知道的事情。而输出意味着“知道”
1/ 在 ChatGPT 中实现了一门新的编程语言:GPTLang,并用这个语言写了一个排序算法。 定义了一个新的命令 `gptlc`,用来编译 GPTLang 的代码。 下图是最终的效果:让 ChatGPT 用 GPTLang 写了一个选择排序,并在命令行编译运行。 这个 thread 将会详细讲述一下是如何一步步实现这门语言的。
1
1
2
196
有些观点就说人要在和ai交互的loop里面不断学习不断提出新要求,说这是未来的人机交互模范。这和一个专门为llm构建结构化语言对立,前者任然停留在用现有自然语言去学习理解然后再执行,后者直接一步到位的“输出”自己的理解或者想法。
1
116
分水岭可能是,未来哪天llm本可以达到某种水平但是因为自然语言的不严谨性所以无法实现,并且llm的发展也达到了上限,这时候可能就是人工语言登场的时候。 很好奇现在prompt engineering发展到哪一步了
1
114
LOL
I would say that we are undoubtedly at AGI when AI can create a real, living unicorn. And no I don’t mean a $1B company you nerds, I mean a literal pink horse with a spiral horn. A paragon of scientific advancement in genetic engineering and cell programming. The stuff of childhood dreams. Dare I say it will happen in our lifetimes
1
124
First off, congrats to these dudes. I still remember one of Roy’s first videos where he gamed that Amazon interview live. It was a marketing masterclass. Now to the hate. Cluely ends up getting a lot of it because it became the poster boy for a shift that was already underway in Silicon Valley. Silicon Valley used to be where the nerds built quietly, fought the big guys, and turned into assholes along the way. People respected that. It was the underdog story. Now you're expected to throw a frat party on day one. Build an audience. Be loud. That’s way outside the comfort zone for most of the old guard. Imagine having to build a Tiktok audience, when half the room thinks it's cringe. No one is going to fund you without that either. And this won’t stop with founders or marketing interns. Next up, even founding engineers and early hires will be expected to have some social media clout before they’re hired. That’s what’s spooking people. We haven’t seen anything yet. Wait till the next YC batch walks in with the same playbook. The hate is because the last bastion of the introverted nerd is falling. And there’s nothing they can do. Since 2021, there’s been too much money floating around in silicon valley. ZIRP, crypto, AI boom. Being a founder is now a status symbol and you can also make money, if you play the cards right. Everyone’s cosplaying the underdog builder, some without even realizing it. That’s why you see the 3AM grind selfies, the mattress-on-the-floor shots, the thread bois apocalypse. And there is nothing much anyone can do either. Cost of starting up or building products has gone down. Whats left is marketing, and here the only option is to get as loud as you can, or be edgy or cringe. Cluely spotted this trend and they are just riding this tide. Good for them.
announcing @cluely's $15M fundraise, led by @a16z. cheat on everything.
57
61
29
1,030
276,118
Saying that human can control itself is a scam. But it isn't means everyone should follow some kind of "internal mind" or "primal desire ", which is a superficial reductionism.
I was one of them. It's selfish, but at the end I had to think about myself. I knew otherwise we would both sink even deeper into the darkness. (We both had/have severe case of depression and anxiety)
1
190
I hated that the notion, emotion themselves and why they are look like this. then I tried to separate from them and found a way out
132
Ziggy Chen retweeted
Companies going AI-first should dedicate some talent that knows what AI is capable of to be in the trenches to design next gen workflows. AI moves so fast it’s hard to decentralize this knowledge yet. Huge opportunity for engineers and IT folks that jump on this early.
We are opening up a new role at Quora: a single engineer who will use AI to automate manual work across the company and increase employee productivity. I will work closely with this person.
43
68
6
483
142,631